Predicting the future has always been one of the objectives of those who do business, perhaps one of the most strategically relevant activities. Those involved in running a business know how complex it is, especially since predicting the future involves trying to understand what happened in the past. However, we all know very well that what happened a while ago is not necessarily going to happen again in the future; on the contrary, in most cases something very different is likely to happen. This is an old diatribe that is destined to remain so. Focusing on the financial sphere, are you familiar with technical analysis? That is, the analysis of the trend of a security to understand what will happen in the immediate future? This is based on an analysis of a so-called historical series, i.e., the stock’s performance over the past to understand trends and behavior and what might happen in the future.
Forecasting, predictive analysis to support the CFO
We are aware, however, that when there is a Board of Directors meeting, perhaps an extraordinary Board of Directors meeting, some important event may occur, both positive and negative for that company that was not foreseeable at the outset and, inevitably, the technical analysis will not have been able to take it into account. In this case the forecast will be missing an aspect, perhaps the most important one, which if not considered could lead to wrong business decisions. That is why this is not a simple area. In short, forecasting in business, but also in any field, is a fundamental decision support.
It is based on mathematical models, and we know that a model, no matter how well done, always presents a partial view of reality. So, the human being must understand this and use the suggestion born from the model but combined with what is happening in the context of real life.
Forecasting, predictive analysis. Mathematical and statistical models
When forecasting we can basically divide the world into two sets: mathematical-statistical models and artificial intelligence methods. The former tends to be used when there is not a lot of data available but there is enough to try and estimate a forecast. What is done in these cases is the famous time series analysis – I have a list of values that refer to a quantity that varies over time and I try to estimate how these values will change in the future, so I can consider seasonal aspects and whatnot. Let’s take an example: if I must make a forecast of the bookings of a hotel, I will certainly have a historical series containing the number of rooms booked day by day. If the objective is to estimate how bookings will go next year, I will have to find an output where for all the days of next year I will have a number that corresponds to the estimated number of booked rooms. To do this, the statistical mathematical method that will be used in this case will certainly have to evaluate seasonal aspects, for example if the location is seaside, I expect a peak of bookings in August especially if we are in Italy. But not only that: also, seasonal aspects within each individual week, for example I expect more rooms occupied at the weekend. And what’s more, if I am lucky enough to have the historical data for several years, if there is a growing trend of bookings, I can expect further growth next year. And here is the big assumption of making predictions on historical data: if next year the marketing department makes a very successful campaign maybe sales increase much more or something happens inside the hotel and customers give a lot of negative reviews, then bookings could plummet. All these events are unforeseeable with a simple analysis of the story. The analyst will come up with a forecast that will obviously not be accurate and, as I said before, he will have a decision support to use as a comparison to what he expects as a human being.
Forecasting, predictive analysis. Artificial Intelligence Methods
Let’s now imagine that we are in the luckiest case, the one where we have a lot of data and maybe even a lot of variables to consider, we always have historical data, but it is more and more workable. In this case, we can try using artificial intelligence. The big advantage over the mathematical-statistical methods I mentioned earlier is that artificial intelligence can find hidden patterns in the data, it doesn’t just analyze seasonality, which is an obvious variable, especially in certain sectors, but it finds others that allow you to build a clearer picture of the conditions that led to generating a certain number of sales and obtain more precise forecasts. This also makes it possible to detect ongoing changes and change the forecast accordingly. So, to put it bluntly, not even artificial intelligence could have predicted covid-19, however, an AI algorithm used there would have realized sooner than others that something was changing.
Forecasting, predictive analysis. The importance of the database
So, the prediction would not have been obvious, but it would certainly have been better than a prediction that was based only on classic seasonal effects and trends over several years.To conclude, algorithms of this type are invaluable for companies, but they must be used with the right awareness. Before developing them, it is as always necessary to prepare your company’s information system for analyzing the data, which means well-structured and accurate data.